Compiler bug: wrong type-bound procedure called

WileyOne — Mon, 10 Feb 2025 17:46:18 GMT

Compiler: ifx (IFX) 2025.0.4

OS: Rocky Linux 8.9 (Green Obsidian)

This is a strange one. The code compiles fine but crashes at runtime; it seems to be an issue with the underlying procedure table. In short, the wrong type-bound procedure is called, eventually leading the program to crash.

The following is a much reduced example from our team's architecture. I expect the type-bound procedure `run` on class `component` to be called, but the example crashes on line 98. As I paired down code from our original situation, the behavior of the program varied at this location; sometimes the obviously wrong procedure would be called, other times it would enter an endless loop until a stack overflow occurred, and still others crashed immediately (like the provided example). It is this behavior that original led me to suspect a compiler bug.

module Component_Header implicit none private public :: component, componentList type :: component contains procedure, public :: run end type type Element class(component), allocatable :: object end type type componentList private type(Element), pointer :: item => null() contains final :: final_componentList generic, public :: assignment(=) => copy_componentList procedure, public :: & add => add_componentList, & get => get_componentList procedure, private :: & copy_componentList end type contains subroutine run(this) class(component), intent(inout) :: this print *, 'this routine should run' end subroutine subroutine final_componentList(this) type(componentList) :: this if (associated(this%item)) deallocate(this%item) end subroutine subroutine copy_componentList(lhs,rhs) class(componentList), intent(out) :: lhs class(componentList), intent(in) :: rhs call lhs%add(rhs%item%object) end subroutine subroutine add_componentList(this,something) class(componentList), target, intent(inout) :: this class(component), intent(in) :: something allocate(this%item) this%item%object = something end subroutine function get_componentList(this) result(something) class(componentList), target, intent(in) :: this class(component), pointer :: something something => this%item%object end function end module module Container_Header use Component_Header implicit none private public :: Container type :: Container ! oddity #1 type(componentList) :: components contains procedure, public :: & runComponents, & addComponent end type contains subroutine addComponent(this) class(Container), intent(inout) :: this call this%components%add(component()) end subroutine subroutine runComponents(this) class(Container), intent(inout) :: this class(component), pointer :: comp => null() print *, 'running the container' comp => this%components%get() call comp%run() !!! this line breaks !!! end subroutine end module module Mode_Header use Container_Header implicit none private public :: mode, mode_constructor type :: ModeInterface ! oddity #2 end type type, extends(ModeInterface) :: mode type(Container) :: myContainer contains procedure, public :: execute end type contains function mode_constructor() result(this) ! oddity #3 type(mode) :: this call this%myContainer%addComponent() end function subroutine execute(this) class(mode), intent(inout) :: this print *, 'execute was called' call this%myContainer%runComponents() end subroutine end module program main use Mode_Header implicit none type(mode) :: myMode myMode = mode_constructor() call myMode%execute() end program

This code is also attached as a file below. Any and all help is appreciated.

Additional Details

Someone might initially be tempted to think that a dangling pointer causes this crash. After all, the `component` object created on line 88 is finalized when exiting the constructor on line 139, so the new `item` pointer existing within in `myMode` is now referencing invalid memory. While it's true that the original `component` object is finalized, the class `componentList` has a defined `assignment` operator - `copy_componentList` - that performs a deep copy. Therefore, the `item` pointer within the `myMode` object references its own copy of `component`. Stepping through the program with gdb confirms that these routines behave as expected.

Note some of the oddities I observed when pairing down our architecture to the given example:

Cutting `Container` and moving its functionality into `mode` causes the error to vanish.
Extending `mode` from `ModeInterface` produces the bug but behaves as expected when `mode` is not an extension of a base class.
Switching the routine `mode_constructor` from a function to a subroutine fixes the error. This is not the case, however, where we originally encountered the bug.

These are not realistic workarounds for our situation; again, this is a significant simplification from our framework. Certain aspects seems silly separated from their context. I am simply explaining why certain features were included in the bugged example.

Re: Compiler bug: wrong type-bound procedure called

WileyOne — Mon, 10 Feb 2025 17:47:50 GMT

Our team also encountered this bug with the following configuration:

Compiler: ifx (IFX) 2024.2.0

OS: Linux Mint 21.3 (Virginia)

Re: Compiler bug: wrong type-bound procedure called

WileyOne — Thu, 26 Jun 2025 17:11:17 GMT

As of 2025.2.0 this is now working. I will wait for confirmation from one of the admins before marking this issue as resolved.

Re: Compiler bug: wrong type-bound procedure called

Ron_Green — Thu, 26 Jun 2025 22:13:22 GMT

Yes, the fix for this bug is in 2025.2.0. I also confirmed the fix just to be sure:

$ ifx -g -O0 -traceback -what -V test.f90 Intel(R) Fortran Compiler for applications running on Intel(R) 64, Version 2025.2.0 Build 20250605 Copyright (C) 1985-2025 Intel Corporation. All rights reserved. Intel(R) Fortran 25.0-1485 GNU ld version 2.35.2-17.el9 $ ./a.out execute was called running the container this routine should run $

Re: Compiler bug: wrong type-bound procedure called

WileyOne — Fri, 27 Jun 2025 11:36:39 GMT

Thank you, @Ron_Green. Please pass along thanks from our team to the developers!

topic Re: Compiler bug: wrong type-bound procedure called in Intel® Fortran Compiler

Compiler bug: wrong type-bound procedure called

Re: Compiler bug: wrong type-bound procedure called

Re: Compiler bug: wrong type-bound procedure called

Re: Compiler bug: wrong type-bound procedure called

Re: Compiler bug: wrong type-bound procedure called